[gate-users] RE: STIR reconstruction software (GATE simualtions)

[Kris Thielemans]

Hi Claude

> 
> > An alternative way would be to let GATE write Interfile 
> headers that 
> > STIR can understand. This shouldn't be incredibly difficult, but I 
> > don't know of anybody who has plans in that direction.
> 
> For now, the sinogram outputs (raw or ecat7) are only defined 
> for ECAT systems as the LOR-to-SinogramBIN conversion follows 
> the ECAT specifications. It means, in particular, that the 
> sinogram radial sampling is not uniform (arc effect) and 
> requires reconstruction routines that handle this specific 
> feature. If we can keep these ECAT specifications, it is 
> quite straightforward to convert an ECAT7 sinogram to an 
> Interfile sinogram (remove the binary headers and directory 
> matrices and write a separate ascii header file, plus byte 
> swapping for the raw data if needed).
> 

You can keep ECAT conventions without changes. 

As an example, STIR contains a utility ifheaders_for_ecat7 which given an
ECAT7 file, it cooks up an Interfile header that points towards the binary
data in the ECAT7 file. Of course, this routine can only give correct
scanner dimensions if the system_type is known.


So, maybe a very ugly work-around would be to let GATE write an ECAT7 file
(with some junk system_type), run ifheaders_for_ecat7 on the data, and then
correct the scanner dimensions etc by hand.

An alternative would be to write some extra code for GATE that essentially
writes only the binary data of the sinogram (just as it would have done for
ECAT7), but writes in addition a fairly simple text file (with the Interfile
header).

Unfortunately, I'm not volunteering.

[In fact, all the stuff that we say is still correct for GE scanners (to the
best of my knowledge, although this is limited...). The main difference
between GE VOLPET and ECAT files is the order of the data and the different
axial compression scheme. In essence, VOLPET uses span=3 for segment 0, and
no axial compression (span=1) for all ring differences>1.]

> An alternative, valid for all PET systems, would be to have 
> an "exact" rebining from the LOR to the sinogram (or 
> projection ) space 
<snip>
> 
> Kris, do you think that such an output would be useful for 
> STIR ?

Not necessary, no. In any case, you wouldn't want that for iterative
reconstructions I think.



Kris